Analysis and visualization with visonerevista-redes.rediris.es/recerca/Egoredes/summer... ·...

Analysis and visualization with visone

Jürgen Lerner

University of Konstanz

Egoredes Summerschool Barcelona, 21.–25. June, 2010

About visone.

Visone is the Italian word for mink. In Spanish visón.

visone is a software for the visual creation, transformationexploration, analysis, and representation of network data.

Outline.

Introducing the visual graph editor.

Analysis and visualization of networks in visone.

Advanced attribute management.

task assignment

Dynamic networks.

Outline.

task assignment

Dynamic networks.

visone’s visual graph editor.

In this trail you’ll learn to use the graphical interface of visoneI how networks can be manually created in visone;I changing the graphical appearance of nodes and edges;I how attributes (e. g., age, gender, type of relationship) can

be added to actors and ties;I using templates and creating a legend;I selecting actors and ties with certain characteristics;I exporting image files suitable for publication.

Networks in visone.

A network in visone encodes three types of information.

I Structure: sets of nodes and edges(who is in the network, who is connected to whom).

I Attributes: properties of actors and ties, includingI external properties such as age, gender, behavior, type of

relationship, . . .I network analytic measures such as centrality, group

membership, . . .

I Graphical appearance: color, shape, coordinates, . . .

Cheat sheet: which file format should I choose?

You have to distinguish between

(I) saving/exporting the whole networkI Whenever possible, use GraphML. That’s the only format which

saves everything (structure, attributes, and graphics).

I Export in Ucinet or Pajek format (dl or net) is supported.I For many other software, including Excel, SPSS, STATA,

R, S-Plus, . . .⇒ adjacency matrices in csv.Note: doesn’t save graphics nor attributes.

(II) saving/exporting an image of the networkI If possible use vector-graphics: pdf, eps, svg.

They give better quality and need smaller file size.

I Otherwise use pixel-based graphics: png, jpg, . . .

Exercise: visone’s visual graph editor.

Draw the social network of (some of) the participants of thiscourse, as far as you are aware of it.

I Include characteristics of the actors.I Include different types of relationships.I Encode the actors’ characteristics in graphical variables.I Don’t forget yourself.I Don’t take this too seriously ;-)

Export an image of your network.

You might do this in groups of two or three to foster discussion.

Outline.

task assignment

Dynamic networks.

In this trail you’ll learn how analysis and visualization work handin hand in visone

I importing networks and attributes from csv tables;I how to layout a network to highlight its structure;I computing network analytic measures such as centrality or

group membershipI how to visualize these;I exporting the computed values in csv tables.

Introduction to the dataset.

Personal networks collected by interviewing ≈ 500 migrants inCatalonia and Florida. (http://www.egoredes.net/)

From each respondent (ego)1. (questions about ego) country of origin, years of

residence, skin-color, health, language skills, . . .2. (alters) 45 people known to the respondent3. (questions about alters) origin, country of residence, . . .4. (ties) which alters know each other

Here, analyze one personal network of a migrant from theDominican Republic to the USA.

Overview: analysis of networks in visone.

visone distinguishes between three different levels of analysis.

1. (indexing) computation of node-level / edge-levelproperties.

I centrality (importance) of nodes and edges;I clustering coefficient;I distance to selected nodes;

2. (grouping) computation of dense subgroups or partitioningthe network into groups.

3. (siena) interface to the RSiena software.I data preparation and model specification in visone;I visualization of the estimated model and its fit to the

observed network;

Cheat sheet: which visualization should I take?

What for?

Layout the network to give a clear picture of its structure.I Usually: use the quick layout button.I For comparison with other software, visone offers other

algorithms (MDS, spring embedder, spectral, circular, . . . ).I Link routing bends edges to reduce crossings.I Label placement reduces overlap.

Highlight node and edge properties.I Attributes (external or network analytic) can be mapped to

graphical variables (size, color, label, coordinates, . . . ).I For displaying importance: centrality and status layout.

Geometric transformation. Rotate, reflect, translate, scale,Procrustes analysis.

What for?

Cheat sheet: importing networks from csv tables.

Import in visone via open (Adjacency Matrix Files).

Take care of setting the correct options.I network type: one-mode vs. two-mode network;I link attribute type: numbers or labels;I row/column labels: present or not (in latter case, actors

are numbered);I cell delimiter: which character indicates a new entry

in the adjacency matrix? (must match export format!);I textframe character to indicate beginning/end of labels

(labels might contain cell delimiter);I merge empty cells ignore repeated cell delimiters

(e. g., two spaces);

Use the file view tab to find out the right options.

Side note: two-mode networks.

Edges connecting actors (objects) of different type.E. g., membership in organizations, written articles, . . .

AUTHORS: A1 A2 A3 A4 A5 A6 A7 A8

ARTICLES: D1 D2 D3 D4 D5

Representation in rectangular matrices.

A1 A2 A3 A4 A5 A6 A7 A8

D1 1 1 1 0 0 0 0 0D2 0 1 1 0 0 0 0 0D3 0 0 1 1 1 1 0 0D4 0 0 0 0 1 1 1 0D5 0 0 0 0 0 1 1 1

Side note: two-mode networks.

Edges connecting actors (objects) of different type.E. g., membership in organizations, written articles, . . .

AUTHORS: A1 A2 A3 A4 A5 A6 A7 A8

ARTICLES: D1 D2 D3 D4 D5

Representation in rectangular matrices.

A1 A2 A3 A4 A5 A6 A7 A8

D1 1 1 1 0 0 0 0 0D2 0 1 1 0 0 0 0 0D3 0 0 1 1 1 1 0 0D4 0 0 0 0 1 1 1 0D5 0 0 0 0 0 1 1 1

Cheat sheet: transformation tab.

Network transformations operate on different levels.

linksI merge putting together parallel links;I split links, e. g., by multiplicity;I simplify, modify direction, . . .

nodesI group by attribute collapsing actors with common

properties;network

I two-mode transformation to one-mode (e. g., co-author,co-citation);

I link graph: ties become nodes;

Exercise: analysis and visualization of networks.

Import of adjacency matrices and attribute tables.I Open the file egoredes/know_each_other.csv

I Import node attributes from the file attributes.csv

Explore the social environment of this migrant.I Display the network structure and simultaneously visualize

different attributes, e. g.,I country of origin (Afrm);I city in which the actor lives (Acit);I type of relationship to ego (Arel).

I Describe the different clusters. Which characteristicsexplain ties in this network?

Who is important?I Compute, visualize, and interpret different centralities.

Outline.

task assignment

Dynamic networks.

In this trail you’ll learn many things about managing attributesI further possibilities to select nodes and links;I types of attributes and their implications;I operations on attributes;I use of non-uniform link-strength and link-length in analysis;I grouping actors by attributes and collapsing groups.

Introduction to the dataset.

NEWCOMB FRATERNITY DATA 15 matrices recording weeklysociometric preference rankings from 17 men attending theUniversity of Michigan in the fall of 1956.

(1) best friend(2) 2nd best friend

. . .(16) least friend

Available, e. g., as a Ucinet dataset, or on the Pajek Website.

Coding of the Newcomb fraternity data.

Data represented in 15 adjacency matrices of order 17× 17 infiles newfrat01.csv to newfrat15.csv

j· 12 2 7

i 1 · 5 103 5 · 86 1 3 ·

Entry in row i and colum j gives the rank of j in the list of i ’sfriends (ordered from the best friend to the least).

Row labels and column labels (actor names) are not included inthe files—but the ordering is the same in all files.

Exercise: advanced attribute management.

Data input and preparation.I open newfrat01.csv;I create a new attribute encoding tie strength;I encode the tie-strength in the thickness of lines.

Popularity of actors.I Compute in-degree weighted by tie strength dependent on

all 16 nominations . . .I and once again dependent on the top-three nominations.I Visualize these two popularity scores in one image.I Does the ordering change? How do you interpret this?

What does it tell you about the actors?

Outline.

task assignment

Dynamic networks.

Task assignment.

Data input and preparation.I Open the file egoredes/egonet_original.graphml

I Look at the edge attributes. Why is this network so dense?I Make an appropriate choice to delete some of the ties.

Explore the social environment of this migrant.I Display the network structure and simultaneously visualize

different attributes (similar to the second exercise).I This time additionally look at attributes coding which actors

are “important” for this migrant (e. g., frequency of contact,“how close do you feel to . . . ?”, etc.)

Save one or more network images with which you can describethe social environment, the acculturation level, etc. of thismigrant. Use these for your presentation on Friday.

Outline.

task assignment

Dynamic networks.

Dynamic networks (recently added features of visone).

Here you’ll learn aboutI visualization of dynamic networks;I an interface to RSiena for modeling network dynamics.

That’s it!

Analysis and visualization with visonerevista-redes.rediris.es/recerca/Egoredes/summer... ·...

Documents

Transcript of Analysis and visualization with visonerevista-redes.rediris.es/recerca/Egoredes/summer... ·...